Time-Sensitive Topic Derivation in Twitter

Much research has been concerned with deriving topics from Twitter and applying the outcomes in a variety of real life applications such as emergency management, business advertisements and corporate/government communication. These activities have used mostly Twitter content to derive topics. More recently, tweet interactions have also been considered, leading to better topics. Given the dynamic aspect of Twitter, we hypothesize that temporal features could further improve topic derivation on a Twitter collection. In this paper, we first perform experiments to characterize the temporal features of the interactions in Twitter. We then propose a time-sensitive topic derivation method. The proposed method incorporates temporal features when it clusters the tweets and identifies the representative terms for each topic. Our experimental results show that the inclusion of temporal features into topic derivation results in a significant improvement for both topic clustering accuracy and topic coherence comparing to existing baseline methods.